NPM and NPM-MLF1 interact with chromatin remodeling complexes and influence their recruitment to specific genes

Nucleophosmin (NPM1) is frequently mutated or subjected to chromosomal translocation in acute myeloid leukemia (AML). NPM protein is primarily located in the nucleus, but the recurrent NPMc+ mutation, which creates a nuclear export signal, is characterized by cytoplasmic localization and leukemogenic properties. Similarly, the NPM-MLF1 translocation product favors the partial cytoplasmic retention of NPM. Regardless of their common cellular distribution, NPM-MLF1 malignancies engender different effects on hematopoiesis compared to NPMc+ counterparts, highlighting possible aberrant nuclear function(s) of NPM in NPMc+ and NPM-MLF1 AML. We performed a proteomic analysis and found that NPM and NPM-MLF1 interact with various nuclear proteins including subunits of the chromatin remodeling complexes ISWI, NuRD and P/BAF. Accordingly, NPM and NPM-MLF1 are recruited to transcriptionally active or repressed genes along with NuRD subunits. Although the overall gene expression program in NPM knockdown cells is similar to that resulting from NPMc+, NPM-MLF1 expression differentially altered gene transcription regulated by NPM. The abnormal gene regulation imposed by NPM-MLF1 can be characterized by the enhanced recruitment of NuRD to gene regulatory regions. Thus, different mechanisms would orchestrate the dysregulation of NPM function in NPMc+- versus NPM1-MLF1-associated leukemia.


Author summary
NPMc+ mutation is the most common mutation in acute myeloid leukemia (AML) with prevalence in one third of all AML cases. NPM can also be involved in leukemogenic translocation including the t(3;5)(q25;q34) NPM-MLF1 translocation, which is associated to bad clinical course but remains poorly defined. We are reporting that NPM and the leukemogenic NPM-MLF1 play central role in chromatin organization and gene regulation in hematopoietic cells. A proteomic analysis provided the evidence that NPM and NPM-

Introduction
Dysregulation of tumor suppressors and oncogenes can influence gene expression either directly or indirectly by altering the structure of chromatin and/or DNA methylation patterns, thereby reshaping the epigenome with consequences for a myriad of cellular functions. This phenomenon is particularly well described in acute myeloid leukemia (AML) [1,2], where characteristic mutations or chromosomal translocations frequently affect the function of transcriptional regulators. For example, the frameshift mutation of NPM1 exon 12 that promotes de novo creation of a nuclear export signal and promotes NPM cytoplasmic accumulation (NPMc+) [3], influences the transcriptome in hematopoietic cells [4,5]. NPMc+ is detected in one third of AML cases and distinctively marks~60% of cytogenetically normal AML (CN-AML; [3]). In addition, NPM1 locus deletion or translocations (e.g., NPM-MLF1, NPM-ALK and NPM-RARα) are associated with many hematological malignancies including AML [6]. NPM1 is one of the most commonly mutated/rearranged genes in hematological malignancies. Determination of the mechanisms underlying the pathological effects of NPM mutation, rearrangement or deletion is complicated by the multiple cellular functions exerted by this factor [7][8][9][10][11]. Evidence for the participation of NPM in gene regulation is mainly supported by results indicating that it can interact with transcription factors [6,12] and cofactors including the chromatin remodeler BRG1 [8] and histone modifiers such as HDAC1 and HDAC2 [10,11]. However, it remains to be determined whether these interactions reflect the protein chaperone function of NPM [9,13,14] or the result of direct or indirect transcriptional regulation elicited by mutant NPM. Similarly, mechanistic details of how mutant NPM might alter transcription regulation remain to be fully elucidated.
The homodimerization capacity of NPM is maintained in NPMc+, thus favoring cytoplasmic localization of the native, non-mutated form of NPM [3]. Since NPM is predominantly located in the nucleus, the cytoplasmic relocalization imposed by NPMc+ creates a deficit in NPM nuclear function thereby promoting leukemia development [15]. This is supported by transcriptomic analyses of bone marrow cells from healthy donors vs AML patients, which demonstrate that multiple genes are deregulated in NPMc+ AML [4,5]. However, the genetic and epigenetic heterogeneity that characterizes AML has hampered the investigation of precise mechanisms responsible for aberrant transcriptional regulation in NPMc+ leukemia cells.
The translocation product NPM-MLF1 retains the capacity to interact with NPM, and the resulting 'heterodimer' partially accumulates in the cytoplasm [16]. Thus, it was proposed that like NPMc+, NPM-MLF1 would promote leukemia through the loss of NPM nuclear function [3]. However, NPMc+ and NPM-MLF1 exert different effects on hematopoietic cells. Unlike NPMc+, which is associated with "frank leukemia" development and is a good prognostic marker, NPM-MLF1 is associated with pre-leukemic MDS (myelodysplastic syndrome) and poor prognosis AML [17]. Given the differential effects on hematopoietic development and the partial nuclear localization of the NPM-MLF1 fusion protein, it is possible that transcriptional programs are altered by a gain-of-function imposed by nuclear NPM-MLF1. Thus, defining the molecular mechanism(s) influenced by the oncoproteins NPMc+ and NPM-MLF1 would have important implications for targeted therapy, particularly for poor prognosis NPM-MLF1 AML.
We sought to determine how NPM and NPM-MLF1 influence the transcriptome of hematopoietic cells and assess whether the effect of NPM-MLF1 expression could reflect the effect of the absence of NPM and/or expression of NPMc+. NPM and NPM-MLF1 proteomic analysis performed in hematopoietic cells led to the identification of novel protein interactions involving NPM and/or NPM-MLF1. Prominent among these interactions were those involving several subunits of the chromatin remodeling complexes NuRD (nucleosome remodeling and deacetylase) [18], BAF, and/or PBAF (Polybromo-associated BAF) [19,20], as well as ISWI (Imitation-SWItch) complexes [21,22], suggesting the direct implication of NPM and NPM-MLF1 in shaping the epigenome of hematopoietic cells. Similar to previous studies of NPMc+ AML patient samples, our transcriptomic analysis indicates that NPM knockdown (NPM KD ) elicits an aberrant gene expression signature. Mechanistically, we demonstrate that NPM promotes the targeted recruitment of NuRD complex to specific genes to regulate their expression, which is significantly impacted by NPM KD . We demonstrate that residual nuclear NPM-MLF1 directly binds chromatin and can enhance recruitment of NuRD to gene regulatory regions. Based on our results, we propose that the NPM-MLF1 translocation contributes to leukemogenesis by the modification of gene expression, which can be associated to the altered recruitment of NuRD.

Characterization of the NPM and the NPM-MLF1 protein interactomes in hematopoietic cells
Western blot assays of Flag-HA-NPM (FH-NPM), FH-NPMc+ and FH-NPM-MLF1 cell lysates indicated that expression of the transgenic proteins was less than 50% of the endogenous NPM protein in all K562 clones (respectively, K562/NPM, K562/NPMc+ and K562/ NPM-MLF1; S1A and S1B Fig), hence precluding artefactual protein-protein interactions due to overexpression. The molecular weight of the FH-NPM and FH-NPMc+ is~40 kDa and that of FH-NPM-MLF1 is~70 kDa. Low transgene expression was also indicated by RT-qPCR (S1C Fig). Since the levels of mRNA are not strictly correlated with transgene protein levels, post-transcriptional mechanisms are likely to be important to control the accumulation of these proteins. The characterization of these models indicated that expression of the transgenes had no significant effect on cell cycle, apoptosis (S1D Fig) or cell proliferation in exponentially growing cultures (S1E Fig), even though the NPMc+ expressing cells escaped growth inhibition when approaching confluence. In any case, during log phase growth, the low-level expression of transgenes does not substantially influence proliferation of the different clones, thus allowing comparative analyses.
Immunostaining detection of FH-NPM, FH-NPMc+ and FH-NPM-MLF1 indicated that, as expected [15,16], FH-NPM was mostly located in the nucleus with strong accumulation in nucleoli; whereas FH-NPM-MLF1 rather than being located to nucleoli, was found in nuclear puncta and exhibited an increased localization in the cytoplasm (S1F Fig). Similarly, a portion of FH-NPMc+ also relocated to the cytoplasm (S1F Fig). This is consistent with the previously described localization of NPMc+ in the patient-derived leukemic cells OCI-AML3 (S1G Fig), which express NPMc+ as a result of oncogenic mutation [23]. Finally, in both NPMc+ and NPM-MLF1 expressing cells, the endogenous NPM was detected in cytoplasmic and nuclear fractions (S1H and S1I Fig).
The prevalent model to explain how NPMc+ and NPM-MLF1 can promote leukemia is related to their capacity to relocate NPM to the cytoplasm and hence, favor nuclear depletion of NPM [3,15,16]. Although the cytoplasmic accumulation of NPMc+ was demonstrated to be required for its leukemogenic capacity [24], NPMc+ and NPM-MLF1 might impose a nuclear gain-of-function due to the displacement of NPM-interacting proteins within the nucleus and/or by modifying cofactor recruitment to specific genes. The latter is particularly conceivable for NPM-MLF1, since MLF1 overexpression and nuclear localization were reported to affect regulation of specific genes [25,26]. To test this hypothesis, we defined the NPM and NPM-MLF1 interactomes in hematopoietic cells. The identification of nuclear proteins interacting with NPM and/or NPM-MLF1 was performed by tandem immunoaffinity purification (TAP-Tag) followed by LC-MS/MS analysis [27,28]. TAP-tag purification was monitored by silver staining (Fig 1A), and the lists of potential interacting proteins are presented in S1 File. NPM and NPM-MLF1 interactomes were analyzed using STRING database (https://www.ncbi.nlm.nih.gov/pubmed/27924014; accessed on 20/08/2018) and the proteins were regrouped according to cellular components and biological processes (Table 1; Fig 1B  and S2 Fig). In agreement with the IF analysis (S1F and S1I Fig), NPM interacts with many nucleolar proteins whereas NPM-MLF1 interacts mostly with nucleoplasmic factors (Fig 1B). Proteins associated with organelles, macromolecular complexes, cytosol, chromatin, and cell junctions are represented in similar proportions in the NPM and NPM-MLF1 interactomes. However, the percentage of proteins in catalytic complexes and cytoskeleton is enhanced in the NPM-MLF1 interactome, and the percentage of proteins involved in RNA splicing is larger in the NPM interactome ( Fig 1B). Furthermore, biological process analysis suggests that NPM is primarily implicated in ribosomal function, consistent with its localization in nucleoli (S1F Fig). On the other hand, the NPM-MLF1 interactome was enriched in nucleoplasmic proteins including those involved in gene regulation. Thus, the comparative analysis of the interactomes of NPM and NPM-MLF1 indicates differences in proteins with nuclear functions and suggests that the oncogenic NPM-MLF1 fusion protein may affect distinct cellular mechanisms due to alteration in its affinity (vs NPM) for specific nuclear complexes. Interestingly, our interactome analysis revealed that both NPM and NPM-MLF1 are associated with chromatin organization factors since many peptides corresponding to subunits of the remodeling complexes NuRD, P/BAF and the ISWI family [18,20,29,30], were identified by LC-MS/MS (Table 1 and S2 Fig). According to the number of peptides and subunits detected for NuRD or P/BAF complex, the NPM interactions with these remodeling complexes are likely to be less frequent or less stable than the corresponding interactions with NPM-MLF1. Indeed, the number of NPM peptides detected by the LC-MS/MS analysis was greater for the NPM TAP-Tag than for the NPM-MLF1 TAP-Tag analysis (Table 1), and no apparent difference was noted for NPM vs NPM-MLF1 interactions with the ATPases SNF2H (SMARCA5) and SNF2L (SMARCA1), which are alternatively included in the ISWI-type complexes [22]. The complete lists of factors involved in transcriptional regulation identified by the LC-MS/MS analysis are presented in S2 File. Many of these potential direct or indirect interactions (hereafter referred to as interactions) were subsequently confirmed by co-immunoprecipitation (co-IP) with Flag, CHD4 (NuRD ATPase) and SNF2H (ISWI ATPase) antibodies, hence further supporting the interactions between NPM-MLF1 and subunits of the NuRD, P/BAF and ISWI-type complexes (Fig 1C and 1D). The co-IP were performed on nuclear extracts that contain deniable traces of DNA (S3 Fig) and DNaseI treatment did not affect NPM and NPM-MLF1 protein with Flag, CHD4, SNF2H antibodies or isotype-matched immunoglobulins (IgG); the antibodies used for Western blots are indicated on the right side of the panels; antibodies were specific for NuRD (panel C) or ISWI and P/BAF (panel D) complex; APE1 was used as negative control; E) Immunoprecipitation assay with Flag antibodies; K562 nuclear extracts of MOCK, NPM or NPM-MLF1 were treated with 10 μg/mL of DNaseI; F) Co-immunoprecipitation experiments were performed with NPM, BRG1, SNF2H antibodies or isotype-matched immunoglobulin (IgG) in HEL WT cell nuclear extracts. The proteins detected by Western blot are indicated on the right side of the panels. interactions with CHD3/CHD4, MTA2, MBD3 ( Fig 1E). Finally, since we also observed these interactions in the erythroleukemia cell line HEL (Fig 1F), we conclude that (i) interactions between NPM and CHD4, BRG1 or SNF2H are not cell line-dependent; and (ii) the endogenous NPM can interact with these remodeling complexes in the absence of the transgenic NPM or NPM-MLF1 protein.
We focused on the interactions of NPM and NPM-MLF1 with NuRD and P/BAF since the TAP-Tag results suggest that the subunits of these two complexes preferentially interact with NPM-MLF1 (vs NPM; Table 1 To further demonstrate complex formation, FPLC was also performed after Flag immunoaffinity purification of FH-NPM and FH-NPM-MLF1 nuclear extracts (Fig 2A). Size exclusion chromatography of Flag immunoaffinity purified proteins indicated that NPM and NPM-MLF1 co-elute with CHD3/CHD4, HDAC1, and MBD3 in different fractions regrouped in three distinct peaks. The NPM-NuRD and NPM-MLF1-NuRD peaks/complexes were regrouped in fractions around 2000 kDa, 750 kDa and 600 kDa (Fig 2A; S5A Fig). Nonetheless, the fractionation pattern of NPM-MLF1 was different from that of NPM since a large portion of the former was included in fractions~2000 kDa to 750 kDa and had a distinctive increase of NuRD subunits and BRG1 in fractions near 1100 kDa (right panels; Fig 2A). Thus, the fusion protein NPM-MLF1 is likely to impose variations in composition of the remodeling complexes NuRD and P/BAF. To gain information regarding the effect of NPM-MLF1 on complex formation, we repeated the immunoaffinity purifications of FH-NPM and FH-NPM-MLF1 ( Table 2 and Fig 2D. The comparative analysis of the MS results indicates that NuRD and ISWI subunits eluted primarily in 'peak 2' fractions of FH-NPM and FH-NPM-MLF1 samples. The subunits of the P/  PAF are primarily in 'peak 1' of the FH-NPM sample but importantly, half of P/BAF subunits detected in the HF-NPM-MLF1 sample are primarily eluting in 'peak 2' instead of peak 1, thus suggesting the reorganization of P/BAF in NPM-MLF1 expressing cells. Based on these results, we next assessed the existence of a complex containing CHD4 and BRG1. Using co-IP procedure, we could not detect interaction between these nucleosome remodeling ATPases in the NPM or NPM-MLF1 cells (S5B Fig), thus suggesting the absence or low occurrence of complex(es) including both remodeling activities. Altogether, the above results suggest that NPM-MLF1 and NPM interact and co-elute with multiple subunits of NuRD, P/BAF and ISWI into complexes of~2000 kDa and~1100 kDa in hematopoietic cells. Furthermore, the size fractionation analysis suggests a reorganization of remodeling complexes in NPM-MLF1 expressing cells.

Identification of wild-type and mutant-NPM-regulated genes: evidence for altered transcriptomes in the etiology of NPM-mutated malignancies
The expression level of multiple genes is affected in CN-AML patient samples characterized by the NPMc+ mutation [4,5]. To investigate the effect of NPM nuclear loss-of-function, as in NPMc+ leukemic cells and identify NPM target genes, the knockdown of NPM (NPM KD ) was performed in K562 cells. Different shRNA were tested and two were selected based on their capacity to reduce NPM levels (shNPM300 and shNPM298; Fig 3A).  Fig and S4 File). The GO term analysis of these genes, generated with DAVID bioinformatics software, indicated that the genes affected by the NPM KD encode proteins involved in various cellular processes ( Fig 3B). Nonetheless, the majority of these genes belong to two processes, namely: (i) cellular adhesion and interaction with the milieu; and (ii) stress related response, including immune response.
To further investigate the transcriptional regulation mechanism(s) influenced by NPM, we selected three genes from the biological processes that were identified ( Fig 3B) and affected by NPM KD in K562 cells (S6C Fig). As they are also abnormally regulated in AML samples characterized by NPMc+ mutation [4,5] we selected: CAspase Recruitment Domain-containing protein 9 (CARD9); Secreted Protein Acidic and Rich Cysteine (SPARC); and Membrane-Spanning 4-domains subfamily A member 3 (MS4A3 or HTm4). CARD9 can activate the NF-κB pathway [31], which is critical to inflammation but also tumor cell survival, migration and proliferation [32]. CARD9 can be abnormally regulated in NPMc+ CN-AML samples (according to data GEO profiles, NCBI from [5]; accession number: GDS4500) [4]. SPARC is downregulated in MDS [33] and upregulated in AML carrying the t(8;12) or inv(16) rearrangement [34] and SPARC overexpression promotes Imatinib resistance in chronic myeloid leukemia [35]. SPARC is an extracellular collagen-binding protein affecting cell interactions with their environment [36]. Finally MS4A3 which is involved in integration of signals from the cell environment [37], was also selected since overexpression of EVI1, a marker of poor prognosis in adult AML, results in MS4A3 repression [38,39].
Transcriptome analysis indicated that CARD9 is downregulated in NPM KD cells ( Fig 3C  and S4 File). SPARC is upregulated in NPM KD cells (Fig 3C; FDR of � 0.05; S3 File), which can also be detected in some NPMc+ CN-AML samples [4,5]. MS4A3 is downregulated in NPM KD cells (S4 File and Fig 3C) as well as in NPMc+ CN-AML samples. Transcriptome results for these genes were validated by RT-qPCR in the shNPM300 and shNPM298 (Fig 3D  and S7 Fig). In NPMc+ expressing cells, expression of CARD9, MS4A3 and SPARC was affected in a similar manner vs NPM KD cells (Fig 3C and 3E). However, the relative expression level of CARD9, SPARC and MS4A3 was clearly different in NPM-MLF1 expressing cells ( Fig  3E). Thus, the effect of NPM-MLF1 on gene regulation differs from the 'nuclear loss-of-function' imposed by NPM KD or by expression of NPMc+.

NPM influences the recruitment of NuRD to target genes
We first assessed whether FH-NPM or FH-NPM-MLF1 could be recruited to the SPARC, CARD9 and MS4A3 genes in hematopoietic cells. ChIP-qPCR analysis indicated that FH-NPM and FH-NPM-MLF1 are recruited to the Transcriptional Start Site (TSS) region of SPARC, CARD9 and MS4A3 (Fig 4A). Next, we questioned whether NPM could influence the recruitment of the NuRD subunit CHD4 to these genes. In shSc cells, CHD4 was detected at the TSS of SPARC, CARD9 and MS4A3. CHD4 recruitment was lost in NPM KD cells (Fig 4B). Upon expression of NPM-MLF1, CHD4 significantly increased at SPARC and CARD9 TSS (Fig 4C). Although not significant, changes were also observed at MS4A3 TSS. Semi-quantitative Western blot analysis revealed higher BRG1 expression in NPM KD cells whereas CHD3/4 levels increased in NPM-MLF1 and NPM KD cells (Fig 4D). Since CHD3/4 upregulation is detectable in both NPM-MLF1 and NPM KD cells, which have an opposite effect on the expression of NPM target genes, it can be assumed that the expression level of CHD3/4 alone is not sufficient to modify the regulation of SPARC, CARD9 and MS4A3 genes.
Together, these results indicate that NPM and NPM-MLF1 can bind the chromatin of either transcriptionally active or repressed genes. Furthermore, NPM-MLF1 can facilitate the NPM-MLF1 affects chromatin remodeling complexes and transcription CHD4/NuRD occupancy to genes deregulated. BRG1 recruitment could not be assessed due to very low signal-to-noise ratio. We suspect that the epitope could be masked by surrounding interacting proteins. However, the analysis of ChIP-seq previously performed in K562 cells (GEO accession: GSM1003510 and GSM935633) suggests that CHD4 and BRG1 can both be recruited to the CARD9, MS4A3 and SPARC genes (S8 Fig).

NPM and NPM-MLF1 influence transcription initiation and elongation
We then investigated the effect of NPM and NPM-MLF1 on chromatin and promoter organization as well as transcription elongation at the NPM-activated CARD9 and MS4A3 genes and the NPM-repressed SPARC gene. Chromatin organization was probed by analysis of histone post-translational modifications by ChIP-qPCR. Specifically, we investigated the enrichment levels of H3K27me3, H3K4me3, H3ac (H3K9ac and H3K14ac), H3K36me3 and H3K79me3 at gene TSS, since (i) NuRD can participate in gene repression with the PRC2 complex [40] and the P/BAF BRG1 can oppose PRC2 activity [20]; and (ii) NuRD and P/BAF regulate gene activation and can participate in transcription elongation [27, [41][42][43]. SPARC TSS region was characterized by H3ac, H3K27me3, H3K4me3 and H3K79me3 marks and only H3K27me3 enrichment levels significantly decreased in NPM KD cells (Fig 5A), suggesting the influence of  shNPM298 cells (S9 Fig). The influence of NPM on PRC2 recruitment is supported by the co-IPs of NPM with the PRC2 subunits EZH2 and SUZ12 (S10 Fig). The enrichment of both H3K27me3 and H3K4me3 at SPARC in shSc cells is consistent with the notion of a bivalent/poised chromatin organization favored by NPM recruitment [44][45][46]. However, the investigation of promoter organization suggests that NPM-NuRD might also preclude efficient transcription pre-initiation complex (PIC) formation at SPARC gene as evidenced by the absence of transcription proficient Ser2-phosphorylated Pol II C-terminal domain (PCTD) and the enrichment of TFIID at SPARC TSS only in NPM KD cells (Fig 5B). Rather H3ac, H3K4me3 and H3K79me3 marks, characterized the chromatin of CARD9 and MS4A3 TSS but no significant H3K27me3 enrichment could be detected (Fig 5A). H3K4me3 and H3K79me3 characterize transcriptionally active genes where productive transcription elongation is associated with activity of the histone methyltransferases MLL/SETD1 and DOT1L [44]. In NPM KD cells, CARD9 and MS4A3 downregulation was accompanied by a significant decrease of H3K4me3 and H3K79me3 levels (Fig 5A), suggesting that NPM recruitment favors CARD9 and MS4A3 productive transcription elongation. The possibility that NPM could have a positive effect on CARD9 and MS4A3 transcription elongation was also supported by: (i) the significant decrease of PCTD levels at CARD9 ORF and MS4A3 TSS in NPM KD vs shSc cells (Fig 5B and S11 Fig); and (ii) the interaction of NPM and NPM-MLF1 with the transcription super-elongation complex subunit RBBP5 [47] and the P-TEFb subunit CDK9 (S12 Fig), which is responsible for Pol II Ser2 phosphorylation and, thereby, for the release of promoter-proximal paused Pol II [48]. As observed in the shSc cells, also in MOCK cells the SPARC TSS was characterized by H3K4me3 and H3K27me3 marks; moreover, H3K4me3 enrichment significantly decreased in NPM-MLF1 expressing cells (Fig 5C). Consistent with RT-qPCR expression data (Fig 3E), no significant difference was noted in chromatin organization of CARD9 TSS between MOCK and NPM-MLF1 expressing cells (Fig 5C). However, in MOCK cells, H3K4me3 and H3K27me3 marked the MS4A3 TSS, and the H3K4me3 enrichment significantly increased whereas H3K27me3 enrichment significantly decreased in NPM-MLF1 expressing cells ( Fig  5C). According to TFIID, Pol II and PCTD recruitment, NPM-MLF1 has no significant effect on the organization of the transcription PIC of SPARC, CARD9 or MS4A3 (Fig 5D). Indeed, the significant increase of TFIID recruitment to CARD9 is non consistent with a modulation of PIC formation since no variation of Pol II or PCTD could be detected to CARD9.
In conclusion, NPM-NuRD and NPM-MLF1-NuRD can participate in both, transcription repression as well as activation. In its role as an activator, the NPM-NuRD complex is likely to be linked to productive transcription elongation, as observed at the transcriptionally active CARD9 and MS4A3 genes. As transcriptional repressor at the SPARC gene, the NPM-NuRD complex can promote bivalent/poised chromatin organization characterized by H3K27me3 and H3K4me3 [45,46]. The NPM-MLF1 recruitment to SPARC enhances NuRD recruitment, which is associated to the disruption of the bivalent/poised chromatin organization and to further repression of this gene.

Discussion
We report that NPM and NPM-MLF1 interact with subunits of the chromatin remodeling complexes NuRD, P/BAF, and with complexes of the ISWI family. The proteomics analysis H3K4me3, H3K27me3, H3K79me3, H3K36me3, Pol II, Ser2-phosphorylated Pol II C-terminal domain (PCTD), TFIID antibodies or isotypematched immunoglobulins (IgG); SPARC TSS: -120bp to +8bp; CARD9 TSS: -99bp to -37bp; MS4A3 TSS: -6bp to +74bp; fold enrichments were relative to the neuronal Pax6 regulatory element, used as internal control; � : p �0.05 obtained by unpaired Student's t test; N.S: Non-significant enrichment compared to corresponding IgG. https://doi.org/10.1371/journal.pgen.1008463.g005 suggests that NPM-MLF1 influences the organization of the remodeling complexes NuRD and P/BAF in hematopoietic cells. Additionally, a direct implication of NPM in gene regulation is demonstrated by (i) NPM recruitment to specific genes concomitant with the binding of NuRD; and (ii) the uncontrolled recruitment of the CHD4/NuRD complex in NPM-deficient or NPM-MLF1-expressing cells. We demonstrate that the non-nucleolar accumulation of NPM-MLF1 directly influences gene expression and can affect recruitment of the remodeling complex NuRD to NPM regulated genes. In contrast, the effect of NPMc+ is similar to that of NPM KD and, hence, related to a nuclear loss-of-function.
The cytoplasmic accumulation of NPMc+ promotes leukemia [24]. Like NPMc+, NPM-MLF1 maintains the capacity to interact with and induces the cytoplasmic relocalization of the non-mutated NPM. This is believed to result in a loss of NPM nuclear function and trigger leukemia development [15,16]. This view is supported by the previous demonstration that inhibition of interaction between NPM and NPMc+ favors hematopoietic cell differentiation and loss of AML cell survival [49]. To elucidate the nuclear functions of NPM and define the effect of NPM-MLF1, we identified proteins interacting with NPM and NPM-MLF1. Protein interactome analysis led to the identification of new NPM and NPM-MLF1 partners, which belong to the remodeling complexes NuRD, P/BAF and complexes of the ISWI family. They are known organizer of the chromatin architecture. We focused on the interactions with NuRD and P/BAF since the MS analysis suggested that their subunits preferentially interact with the oncoprotein NPM-MLF1 (vs NPM; Table 1) and hence, could explain gene deregulation associated with NPM-MLF1 expression. This preferential interaction was not observed with SNF2H and SNF2L, the nucleosome remodeling ATPases alternatively included in the ISWI complexes [22].
The literature is relatively silent with respect to any role(s) for NPM in chromatin structure/remodeling. Analysis of the NPM protein interactome in hepatoma cells [50] as well as a high-throughput proteome analysis made in HeLa cells [51] indicated that in addition to interactions with proteins affecting various cellular mechanisms, NPM interacts with histones and a few factors involved in gene regulation. The high-throughput analysis made in HeLa cells identified the ISWI complex subunit SNF2H as a potential interacting partner of NPM [51]. As many as 14 different ISWI complexes have been described [22]. Our protein interactome and co-IP results demonstrate that in fact, NPM and NPM-MLF1 can interact with different subunits (SNF2H, SNF2L and BAZ1B/WSTF; Fig 1D) specific to distinct ISWI complexes. Whether NPM and NPM-MLF1 can be recruited to chromatin with different ISWI complexes remains unknown. However, the unique composition of different ISWI complexes could be required for the participation of NPM (and NPM-MLF1) to several nuclear mechanisms such as transcriptional regulation, DNA repair, and DNA replication.
The interaction of NPM with BRG1 was reported in APL cells [8] where it was demonstrated that the translocation product PML-RARA favors the formation of an atypical transcriptional repressive complex composed of NPM/TOP2B/BRG1 [8]. Our proteomic results indicate that NPM also interacts with BRG1 in hematopoietic cells that are not characterized by the PML-RARA translocation. Interestingly, P/BAF can recruit TOP2A and TOP2B enzymes (both identified in the MS analysis of the NPM-MLF1 interactome; S1 File) to facilitate DNA decatenation and heterochromatin remodeling [52,53]. Thus, NPM and NPM-MLF1 might be associated with TOP2 enzymes when interacting with P/BAF.
Our results on the effects of NPM KD and NPM-MLF1 expression prompted us to identify genes abnormally regulated under these conditions. Specifically, we found that NPM and NPM-MLF1 could bind the chromatin of deregulated genes and influence the recruitment of CHD4/NuRD. Their targeting to genes is most likely mediated by specific transcription factors and some of them, e.g. YY1 and CTCF, were identified by MS interactome analysis (S2 File).
However, when directly assessed by protein co-IP, we observed that both NPM and NPM-MLF interact similarly with YY1 and CTCF (S13 Fig). It remains to be defined whether other transcription factors or DNA structures [54] could differentially interact with NPM vs NPM-MLF1 and induce the preferential nuclear localization of NPM-MLF1 over NPM. Nonetheless, the enhanced recruitment of CHD4/NuRD imposed by NPM-MLF1 is likely to be critical for the leukemogenic activity of this fusion oncoprotein. The influence of NPM and NPM-MLF1 on P/BAF recruitment was not directly tested, but the comparative analysis of published ChIP-seq results obtained in K562 cells (S8 Fig), suggests that CHD4 and BRG1 can both be recruited to the NPM-activated genes CARD9 and MS4A3, as well as to the NPMrepressed gene SPARC. Whether CHD4 and BRG1 could be simultaneously recruited to these NPM regulated genes could not be clarified. Indeed, the proteomic analysis performed was not conclusive since, the size-exclusion chromatography suggests that when interacting with NPM (or NPM-MLF1), CHD4 and BRG1 could be in the same complex and elute in same fractions of size exclusion chromatography, but CHD4 and BGR1 co-immunoprecipitation could not be detected in the K562 cells.
The recruitment of CHD4 and BRG1 to the same genes has been reported by different groups. Indeed, these nucleosome remodeling activities were reported to be associated and involved in the repression of γ-genes during β-globin gene switching [55,56]. Alternatively, the recruitment of NuRD and BAF complexes to the same genes was reported during epithelial-mesenchymal transition in oral cancer cells [57]. Then, NuRD was found to antagonize BAF complex activity during the transition from epithelial to mesenchymal, suggesting that the recruitment of NuRD and BAF to common genes could facilitate the switching between transcriptional activation and repression. NuRD and BAF also show similar chromatin binding in embryonic stem cells [55,58,59]. Thus, such as proposed in embryonic stem cells [60], these examples suggest that the potential opposing activity of CHD4/NuRD and BRG1/BAF could facilitate the fine-tuning of gene expression and/or their capacity to respond to activation or repression signals [58].
NuRD is a critical regulator of chromatin organization within gene regulatory regions and, thereby, can facilitate different events leading to transcription initiation. Importantly, as demonstrated for certain ISWI complexes [60,61], NuRD also influences transcription elongation [27,41]. One of our goals was to understand how NPM and NPM-MLF1 act with NuRD to promote transcriptional activation or repression. Our results indicate that NPM-NuRD is required for the transcriptional repression of SPARC and activation of CARD9 and MS4A3. In the case of SPARC, NPM-MLF1 favors NuRD recruitment with consequent decrease of gene expression. Enhanced recruitment of TFIID at the TSS in NPM KD cells suggests that NPM-NuRD negatively regulates PIC formation at SPARC. Additionally, difference in H3K27 methylation suggests that when implicated in gene repression, NPM-NuRD (or NPM-MLF1-NuRD) act with the PRC2 methyltransferase EZH2. The chromatin and PIC analyses suggest that NPM-NuRD favors a bivalent chromatin characterized by H3K4me3 and H3K27me3, and the NPM-MLF1 recruitment disrupts this chromatin organization to further repress SPARC transcription. Thus, NPM and NPM-MLF1 regulate SPARC transcription by controlling the chromatin organization and consequently, transcriptional activation.
With respect to CARD9 and MS4A3, NPM KD does not influence TFIID recruitment but significantly affects H3K4me3, H3K79me3 and PCTD enrichment, suggesting that transcription elongation but not PIC formation or transcription initiation would be impaired by NPM KD [27]. Indeed, H3K4me3 and H3K79me3 are formed by the histone methyltransferases MLL/ SETD1 and DOT1L, respectively [62], and these chromatin marks can be associated with the transcription elongation complex thus promoting productive transcription elongation. The elongation complex including the MLL/SETD1 and DOT1L activities also contains P-TEFb [47,63] and NPM was reported to interact with HEXIM, a subunit of the repressed form of P-TEFb [64]. The NPM-HEXIM interaction was not detected in K562 cells (LC-MS/MS analysis; S1 File). However, NuRD and P-TEFb are found in a larger complex that facilitates transcription elongation [27,41] and NPM as well as NPM-MLF1 co-immunoprecipitate with the P-TEFb catalytic subunit CDK9 (S12 Fig). Thus, the positive influence of NPM-NuRD on transcription elongation of CARD9 and MS4A3 suggests that NPM-NuRD could act with P-TEFb at these genes. The absence of variation of CARD9 expression in MOCK and NPM-MLF1 expressing cells could suggest that NuRD recruitment to CARD9 is optimal under normal conditions, and the enhanced recruitment of NuRD in NPM-MLF1 expressing cells would not further promote transcription elongation at this gene. Importantly, CARD9 is a NF-κB inducible gene [31] and may require the binding of this stress-induced transcription factor to be further activated. Thus, the NPM-NuRD effect on CARD9 and MS4A3 transcription is possibly related to improvement in transcription elongation since NPM interacts with CDK9/ P-TEFb, and it favors MLL/SETD1 and DOT1L activities. It is not clear at present whether NPM-MLF1-NuRD can similarly affect transcription or acts more specifically on the chromatin organization at promoters.
In summary, the mutation of subunits of ATP-dependent chromatin remodeling complexes can favor leukemia [65][66][67]. Our study provides evidence that NPM and the leukemogenic translocation NPM-MLF1 interact with the chromatin remodeling complexes NuRD, P/BAF as well as complexes of the ISWI family. We show that NPM and NPM-MLF1 can both recruit and promote the capacity of the CHD4/NuRD complex to regulate gene expression in hematopoietic cells. NPM nuclear deficiency results in the abnormal regulation of NPM-and CHD4/ NuRD-target genes. The oncoprotein NPM-MLF1 maintains the capacity to interact with NPM and is partially located in the cytoplasm. Nonetheless, unlike the nuclear 'loss-of-function' imposed by NPM KD or NPMc+, the nuclear NPM-MLF1 binds to the chromatin, influences the organization of remodeling complexes, favors the uncontrolled recruitment of NuRD and induces abnormal transcriptional regulation of NPM target genes. The balance between remodeling complex NuRD and BAF is critical for the regulation of genes [57]. We propose that NPM-MLF1 disrupts the balance between NuRD and P/BAF complex to certain genes and thereby, induces their abnormal regulation. The results presented here suggests that NPM-MLF1 is associated to nuclear gain-of-function, rather than by a previously proposed loss of nuclear function effect, such as imposed by NPM KD or NPMc+. The discovery of new protein partners of NPM and NPM-MLF1 and demonstration of mechanisms used by these factors to regulate genes provide a new basis for our understanding and further investigation of the NPM-related oncoprotein driven AML and their poor prognosis, especially those carrying the oncogenic fusion NPM-MLF1.

Nuclear extraction and co-immunoprecipitation assays
Nuclear extracts were prepared as previously described [68]. Briefly, cells were incubated for 10 min on ice in a hypotonic buffer (10 mM Tris pH 7.3, 10 mM KCl, 1.5 mM MgCl 2 ). Cells were then lysed using a dounce homogenizer and isolation of nuclei was monitored under the microscope. After centrifugation, the nuclear pellet was resuspended in two equivalent volume of low salt buffer (20 mM Tris pH 7.3, 25% Glycerol, 0.2 mM EDTA, 20 mM KCl, 1.5 mM MgCl 2 ). One volume of high salt buffer (20 mM Tris pH 7.3, 25% Glycerol, 0.2 mM EDTA, 1.2 M KCl, 1.5 mM MgCl 2 ) was then added dropwise in the cold room. Following extraction of nuclear protein, chromatin was separated by ultracentrifugation (15000 rpm, 30 minutes and then 20000 rpm, 20 minutes at 4˚C). All buffers were completed with 1% protease inhibitor cocktail (Bimake, Catalog Number: B14001), 1 mM PMSF, 10 mM β mercapto-ethanol. For Co-immunoprecipitations (co-IP), the nuclear extract was centrifuged at 13000 rpm for 10 minutes at 4˚C and the supernatants were diluted in the equivalent volume of IP buffer (50 mM Tris pH 7.5, 10% Glycerol, 0.2 mM EDTA, 145 mM KCl, 5 mM MgCl 2 , 0.5% NP40, 1% protease inhibitor cocktail (Bimake), 1 mM PMSF, 10 mM β mercapto-ethanol). Specific antibodies or isotype-matched immunoglobulin controls (IgG) (2 μg) were added and lysates were incubated overnight at 4˚C. Complexes were precipitated with protein G-Sepharose beads (P3296, Sigma) or Flag agarose beads (A2220, Sigma) and washed three to six times with 1 ml of ice-cold IP buffer. The antibodies used are listed in S1 Table.

Sequential immunoaffinity purification (Tap-Tag) and Fast protein liquid chromatography (FPLC)
Flag-HA sequential immunoaffinity purification was carried out as previously described [68], from 20 mg of nuclear lysates of K562 (MOCK, NPM and NPM-MLF1) cells. Immunoprecipitated proteins were eluted from Flag or HA beads with 4 μg/ml of Flag or HA polypeptides (sequences: Flag: DYKDDDDK or HA:YPYDVPDYA from BioBasic) and molecular size chromatography was performed on the immune-affinity purified complexes with the AKTA purifier FPLC system (GE Healthcare) using Superose 6 increase 10/300 GL column (GE Healthcare). The column was calibrated with Dextran (2000 kDa), Thyroglobulin (669 kDa), Aldolase (158 kDa) and Bovine Serum Albumin (67 kDa) in FPLC buffer (50 mM Tris pH 7.3, 10% Glycerol, 60 mM KCl, 0.2 mM EDTA pH 8.5 mM MgCl 2 , 1% NP40 and 0.5 mM DTT). 500 μL of nuclear lysate or Tap-Tag purified sample was loaded on the column. After 0.2 column volume of isocratic elution, 0.5 mL fractions were collected, and proteins were precipitated with 15% TCA. Immunopurified samples or fractions, were first analyzed by silver staining, and then LC-MS/MS analysis was performed to identify interacting partners (Taplin Mass Spectrometry Facility, Harvard University, Boston).
Western blots were performed according to standard protocols. The antibodies used are listed in the S1 Table. Images were acquired with Image Quant LAS400 device (GE Healthcare Life Sciences).

Chromatin immunoprecipitation and quantitative PCR analyses
Chromatin immunoprecipitation (ChIP) and quantitative real-time PCR assay were performed as previously described [56]. Briefly, 1 x 10 6 cells were fixed with 1.5 mM ethylene glycolbis[succnimidyl succinate] (EGS; MJS Biolynx Inc), 1% formaldehyde (Sigma) and quenched with 50 mM Glycine (Sigma). The antibodies used are indicated in S1 Table. About 10% of immunoprecipitated material and 3% of input sample were used as template for qPCR with SYBR green (Bimake) on a 7500 Real time PCR System (Applied Biosystems). Primer sequences are described in S2 Table.

Quantitative reverse transcription PCR (RT-qPCR)
RNA was extracted with TRIzol (Invitrogen) and treated with DNaseI-RNase-free (Thermo Scientific). Reverse transcription reactions were performed with oligo(dT) 18 primers (Thermo Scientific) and RevertAid reverse transcriptase (Thermo Scientific), according to manufacturer's instructions. RT material (1/20) was used as template for qPCR with SYBR green (Biotool) on a 7500 Real time PCR System (Applied Biosystems). Primer sequences are described in S2 Table. Quantification was carried out as previously described [56].
Immunofluorescence 5 x 10 5 cells were cytospun on slides and fixed with 3% paraformaldehyde for 30 minutes. Slides were washed with immunofluorescence (IF) wash buffer (0.1% Triton, 1 mM Sodium Azide in PBS) and blocked with 5% Bovine Serum Albumin for 1 hour. Cells were first stained with rabbit anti-HA and mouse anti-NPM and then with Alexa Fluor 594-conjugated anti-rabbit and Alexa Fluor 488-conjugated anti-mouse antibodies (S1 Table). Slides were mounted with DAPI-Dabco:Vectashield solution (1:1). Images were acquired with Delta Vision Core/ Olympus IX71 microscope and were deconvoluted at 200 nm with SoftWoRx 6.5.2 software.
Cell cycle analysis 2 x 10 6 cells were washed with cold PBS, fixed with 75% cold ethanol and incubated overnight at -20˚C. Cells were then washed with PBS and treated with 100 μg/mL RNase A (Sigma) for 20 minutes at 37˚C. DNA was stained with 50 μg/mL Propidium Iodide. Cell cycle analysis was carried out on a FACScan flow cytometer with CellQuestPro software (BD Biosciences) according to the DNA content [69].

Ampli-sequencing assay
cDNA libraries were prepared from 10ng of K562 mRNA pre-treated with DNaseI, in triplicate, with the Ion AmpliSeqTM Transcriptome Human Gene Expression Kit (Thermo Fisher Scientific) according to manufacturer's instructions. Briefly, mRNA were reverse transcribed using SuperScript VILO cDNA synthesis kit (Thermo Fisher Scientific) and amplified with Ion AmpliSeq HIFI Mix. Primer sequences were partially digested with FuPa Reagent, and then barcoded using Ion Xpress Barcodes (Thermo Fisher Scientific). Purification was carried out by AMPure XP Reagent (Beckman Coulter). Libraries concentrations were defined by qPCR using Ion Library Quantification kit (Thermo Fisher Scientific). Libraries were pooled together at 50 pM for emulsion PCR, carried out using the Ion Chef Instrument (Thermo Fisher Scientific). Purified Ion Sphere Particles were loaded on Ion P1 Chip. Loading density was around 80%. The sequencing was performed on Ion Proton system (Thermo Fisher Scientific). Ion Torrent software, Torrent Suite v4.4 (Thermo Fisher Scientific) was used for base calling, alignment to the human reference genome (hg19) and quality control. Raw reads were then analyzed automatically using the AmpliSeqRNA plugin to generate gene-level expression values for all 20802 RefSeq human genes.

Statistical analysis
Unpaired Student's t-test was used to determine the level of statistical significance (P-value).